@berryxia

我靠，Vibe Coding 太特么上头了… 但是，更上头的是我这该死的“强迫症” 下面这个片子，本来是我已经初步剪好的一个正片视频 demo 演示，但最终还是被我干掉了。 因为我发现还有一些细节内容需要打磨和提升，真的是绝了。 这个东西已经改了两个礼拜还没搞完，其实核心功能都已经实现了。救救我吧，兄弟们！ 文旅局的朋友们可以看过来，这是一个非常好玩的东西。 能不能开源，就取决于你们的热情度了。好了！ Video

x.com/i/article/205852961337…

The best VLLM scores only 14% on oracle bone script recognition. Chronicles-OCR, a new ancient Chinese character benchmark from Tencent HY and 4 institutions, just put 28 frontier models to the test across 3,000 years of Chinese writing. 🏺🗂️ modelscope.cn/datasets/Virtu… 2,800 expert-annotated images across 7 scripts: oracle bone, bronze, seal, clerical, regular, running, and cursive. Three findings worth knowing: 🔍 End-to-end detection collapses on ancient scripts. Best H-mean: 16.5. GPT-5 and Gemini 2.5 Pro near 0. 🧠 Reasoning makes it worse. Thinking mode hurts performance across almost all models — chain-of-thought amplifies hallucination when perception fails. 👁️ Models recognize the carrier, not the script. Ancient script classification hits 96.7% by spotting turtle shells and bronze vessels, not the characters themselves.

x.com/i/article/205957789644…

Today we’re announcing our $113M Series B led by @CapitalGVC. Over the last 6 months, weekly volume on OpenRouter grew from 5T to 25T tokens as AI rapidly shifts from experimentation into production. We’re excited for what comes next.

Recently, we took time to consolidate all of the work behind M2 and published it here: our M2 paper on arXiv It’s been just over six months since we first open-sourced M2 on December 23 last year. During that time, a number of our ideas and systems have been broadly adopted by the open-source community — including CISPO, Forge RL System, Self-Evolution. Over the past six months, we’ve felt incredible enthusiasm from the open-source community. Nearly every model release reached the #1 spot on the Hugging Face leaderboard. Now it’s time for a new chapter. We’re getting ready for M3. MSA paper is on the road. arxiv.org/abs/2605.26494

说好不熬夜的，但 AI Coding 太上瘾！ 昨晚开发了个 Chrome 新窗口插件，超方便。 1. 番茄钟、音乐播放、Todo、便签、天气、换背景等，独立开发者多件套整合到了一起 😂 2. 支持谷歌搜索，ChatGPT跳转官网带提示词发送。 3. 支持Command + K唤起，快速设置、搜索一切。 已开源，见评论区。 Video

Typeless 1.5.0 is live for macOS & Windows! ✨ Bringing custom shortcuts to external keyboards. ⌨️ Your favorite setup, working your way. 💙 Go seamless: ⚡️ typeless.com/ #Typeless

iPhone 上直接用 App Store 下载就行： 👉 Bonsai Studio — PrismML 官方 iOS 应用，免费安装，模型在手机本地跑 我觉得给学校老师做一些素材展示，或者幼儿园小朋友教学还是可以的。 不需要额外的TOKEN费用，风格支持的挺多的。 中文文字还是乱码但是可以快速理解你的意境（图2） 技术背景： Bonsai Image 4B 基于 FLUX.2 Klein，把模型权重压缩成 1-bit/3-bit，体积从 7.75GB 压到 0.93GB iPhone 上生成一张 512×512 图约占 1.5GB 内存， 1024×1024 大概 2GB，iPhone 15 Pro 以上没问题 完全本地推理，不联网 就可以跑起来！ Android 暂时没官方 App，只能走 WebGPU 网页版。 我实际在iPhone 17 Pro Max 测试了一下，出一张5125*512的画的速度不到几十秒就可以出来一张。 浏览器中需要下载1.8G 左右模型就可以玩~ 地址在评论区👇🏻

兄弟们，端侧生图模型又有新东西了！ 之前有客户端Drawthing 客户端也是支持iPad 手机进行文生图～ 今天PrismML也来了一个好玩的… 他们发布了1-bit和Ternary版本的Bonsai Image 4B扩散模型。 1-bit那版只有0.93GB，比全精度模型小8.3倍。 Ternary版1.21GB，用-1、0、+1的三元权重，在保持极致小巧的同时，把图像质量和提示词遵循度拉得更高。 两者在Mac M4 Pro上生成速度最高能快5.6倍。 更重要的是，质量上它跟那些大得多的模型打得有来有回。 对象构图、人像偏好、美学评分、复杂提示跟随，全都不落下风。 他们还同步上线了Bonsai Studio这个iOS App。 直接在iPhone上本地生成图像，不用订阅，不用调用API，彻底离线可用。 这套极致压缩技术，把以前只有云端才能玩的高质量图像生成，真正塞进了个人设备。 完了测试效果看看如何… 等我

Something BIG is coming

Gemma 4 meets retro gaming! 🕹️✨ Introducing AIventure: an open-source dungeon crawler designed as a developer masterclass. Learn to build agentic workflows and integrate vibe-coding, like prompting an NPC to build a web app on the fly. Video

Just released oMLX v0.3.11. The last few updates have focused on stability as my top priority, and I'm planning to ship new features, including pending PRs, in the next release. (Huge thanks to everyone who's sent support and contributions!) If you've run into stability issues with oMLX under high memory pressure, I'd really appreciate it if you gave this version another test. github.com/jundot/omlx/relea… As a solo developer with a main job on the side, maintaining a project that's grown bigger than I expected isn't always easy, but getting to grow it together with the open source community has been a genuinely joyful experience. oMLX aims to deliver a Local AI experience that even my friend who bought a MacBook just yesterday can pick up and use right away. I hope oMLX makes it easy for everyone to run Local AI on their Mac. Is there a specific new feature you'd love to see in oMLX? Feature suggestions are always welcome!

Today we’re releasing 1-bit and Ternary Bonsai Image 4B. A new family of image-generation models designed to run high-quality diffusion inference on local hardware: from laptops to phones.

Anthropic 黑客松冠军，把自己的 AI 编程工作台开源了。 项目叫 ECC。作者 Affaan Mustafa 和队友用 Claude Code 8 小时做出 zenith.chat 拿了冠军，后来把背后的工作流整理成这个仓库。 它不是一个提示词合集，里面包含 61 个 Agent、246 个 Skills、76 个预设命令，还带规则、Hook、安全扫描和 MCP 配置。 如果你每天重度使用 Claude Code、Cursor、Codex，这个项目值得翻一遍。 它展示的不是怎么问 AI，而是怎么给 AI 搭一套能持续干活的工作台。 github.com/affaan-m/ECC

x.com/i/article/205928765533…

very cool research (and nomenclature)

卧槽… 兄弟们 我真的是后知后觉～ 今天还屁颠屁颠跑到Bloome 去问什么时候可以支持iOS端啊？ 结果告诉我说美区已经可以使用了… 因为我一直以为没有在iOS 添加到主屏幕在用，忍受着… 结果下载完使用起来太丝滑了…… 真特么好用…

Mark Zuckerberg says Apple's lack of innovation since the iPhone will lead to its decline "They haven't really invented anything great in a while. Steve Jobs invented the iPhone, and now they're kind of sitting on it 20 years later" "Year over year, I'm not even sure they're selling more iPhones at this point. Part of it is that each generation doesn't actually get that much better, so people are taking longer to upgrade" "They built stuff like AirPods, which are cool, but they've thoroughly hamstrung the ability for anyone else to build something that can connect to the iPhone" "I'm pretty optimistic that because they've been so off their game in terms of not really releasing many innovative things... eventually they'll get beat by someone" Video

兄弟们，上次我分享了SenseNova Skills Agent 免费可以领取Token Plan 后，自己也拿来实战了一把。 此项目已经在 Github 斩获 2400Star⭐️ 我让它跑了个AI大模型市场调研报告，从开始到结束差不多花了两个小时。 报告出来是Markdown格式，我又让它转成PDF，最后直接做成了PPT。 今天我专门录了个完整的使用教程，从安装到生成全流程，保姆级那种。 现在还是公测期，完全免费，整体体验还不错。 它也支持多轮迭代，如果网络不稳或者请求太多，直接回复它，它就会继续帮你完成。 这个视频里我接入的是Hermes Agent ，大家有需要可以去试试。 我最喜欢的就是可以直接做调研报告，尤其做国内的一些报告的内容还不错。 限时限免，搞点小任务，也是不错的 感兴趣的直接去官网进行体验吧~· 免费Token Plan 领取地址见评论区~👇🏻 Github 地址：github.com/OpenSenseNova/Sen… Video

兄弟们，上次我分享了SenseNova Skills Agent 免费可以领取Token Plan 后，自己也拿来实战了一把。 此项目已经在 Github 斩获 2400Star⭐️ 我让它跑了个AI大模型市场调研报告，从开始到结束差不多花了两个小时。 报告出来是Markdown格式，我又让它转成PDF，最后直接做成了PPT。 今天我专门录了个完整的使用教程，从安装到生成全流程，保姆级那种。 现在还是公测期，完全免费，整体体验还不错。 它也支持多轮迭代，如果网络不稳或者请求太多，直接回复它，它就会继续帮你完成。 这个视频里我接入的是Hermes Agent ，大家有需要可以去试试。 我最喜欢的就是可以直接做调研报告，尤其做国内的一些报告的内容还不错。 限时限免，搞点小任务，也是不错的 感兴趣的直接去官网进行体验吧~· 免费Token Plan 领取地址见评论区~👇🏻 Github 地址：github.com/OpenSenseNova/Sen… Video